view-team-session
Pass
Audited by Gen Agent Trust Hub on Mar 7, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The script
scripts/generate.pyscans for and reads JSONL files within~/.claude/projects/. These log files contain the complete record of user and agent interactions, which may include sensitive information shared during the session. - [COMMAND_EXECUTION]: The script uses the
webbrowsermodule to automatically launch the generated HTML file in the user's default browser after processing. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by rendering untrusted data from session logs into an HTML viewer.
- Ingestion points: JSONL files located at
~/.claude/projects/*/*.jsonl(referenced inscripts/generate.py). - Boundary markers: None; the logic parses and includes all conversation events.
- Capability inventory: The skill can read local files, create directories and files in the
.claude/output/directory, and launch system processes (browser). - Sanitization: The file
assets/template.htmlimplements a custom Markdown parser that uses basic HTML escaping but fails to sanitize URL protocols in Markdown links, creating an XSS vulnerability wherejavascript:URIs can be executed.
Audit Metadata